CN104680516A - Acquisition method for high-quality feature matching set of images - Google Patents
Acquisition method for high-quality feature matching set of images Download PDFInfo
- Publication number
- CN104680516A CN104680516A CN201510010357.2A CN201510010357A CN104680516A CN 104680516 A CN104680516 A CN 104680516A CN 201510010357 A CN201510010357 A CN 201510010357A CN 104680516 A CN104680516 A CN 104680516A
- Authority
- CN
- China
- Prior art keywords
- coupling
- matching
- point
- polar curve
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 97
- 239000011159 matrix material Substances 0.000 claims abstract description 50
- 230000008878 coupling Effects 0.000 claims description 136
- 238000010168 coupling process Methods 0.000 claims description 136
- 238000005859 coupling reaction Methods 0.000 claims description 136
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 14
- 230000006978 adaptation Effects 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 6
- 238000001514 detection method Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 3
- 238000000746 purification Methods 0.000 claims description 2
- 238000001914 filtration Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 2
- 238000012216 screening Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 6
- 230000032683 aging Effects 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 230000000052 comparative effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 230000013011 mating Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an acquisition method for a high-quality feature matching set of images. For solving the problem that high-quality feature matching numbers obtained by an original method are fewer, so the application demand cannot be met, a novel acquisition method for high-quality matching is provided. According to the method, the matching quality can be improved, and the high-quality matching quantity of feature points is increased. The acquisition method generally comprises the following steps: firstly, detecting, describing and matching image feature points; secondly, obtaining a basis matrix among images to be matched; thirdly, calculating polar lines of feature points on the images to be matched; fourthly, filtering wrong matching by a symmetric filter for the first time; fifthly, performing epipolar constraint, and screening and filtering the matching set by using the relation between the polar line and the feature points on the corresponding images to obtain an excellent matching set; sixthly, removing the matching in which matching points are mutually intersected on the same polar line to obtain a final high-quality matching set. The acquisition method disclosed by the invention is good in effect and is suitable for various image data.
Description
Technical field
The present invention relates to the technical field of images match in computer vision and perspective geometry, particularly a kind of image quality features set of matches acquisition methods based on epipolar-line constraint.
Background technology
In computer vision field, picture point feature can be used for finding the sparse set of corresponding point position in different images, and the result that this some corresponding method obtains is the prior step calculating video camera attitude.From early stage stereoscopic vision coupling, this point correspondence is just widely used.Afterwards, promoted on a large scale in the object tracking process of image mosaic application, automatically three-dimensional modeling, object identification and video sequence again based on a corresponding method.Said method comprises two classes, and one is in piece image, find those local search approach can be used to carry out the feature (that is: based on the method for local pixel area image coupling) of accurate tracking.Two is detect unique point independently in the image of all investigations, and then apparent the carrying out in local based on it mates (that is: the method for feature based Point matching).The former at image with more suitable during close viewing angles, time the latter is more suitable for and in most of the cases there is a large amount of motions or apparent change in image, such as when spliced panoramic figure, when setting up corresponding relation in wide Baseline Stereo vision or when carrying out object identification.Content of the present invention, on the basis of Equations of The Second Kind method, proposes the method that epipolar-line constraint obtains quality features Point matching.
In the method for feature based Point matching, according to the application requirement of target following, image mosaic or object identification, the situations such as the image characteristic point Corresponding matching that algorithm detects must can process that appearance changes, brightness changes, blocks, translation rotation and mirror-reflection.But, because of the above-mentioned change that image exists, in the coupling set of the image matching technology acquisition of current distinguished point based (FAST, SIFT, SURF, ORB, FREAK etc.), there is a large amount of poor quality couplings.The inaccurate of matching result will cause the problems such as such as BREAK TRACK, image mosaic deformity, object identification mistake.A large amount of erroneous matching can be weeded out by introducing arest neighbors ratio strategy and symmetry strategy, re-using stochastic sampling unification algorism (that is: RANSAC), can more reliably matching image feature, improve quality of match.But based on arest neighbors ratio strategy (that is: NNDR) with based in the RANSAC algorithm of disparity constraint, the too low meeting of optimum configurations causes more erroneous matching, and optimum configurations is too high will cause number of matches very little.The appearance of two kinds of situations all can not meet the images match application demand of universality.Based on this, the present invention proposes the image quality features set of matches acquisition methods based on epipolar-line constraint, problem above can be solved well.
Summary of the invention
The object of the invention is the acquisition methods proposing a kind of image quality features set of matches, and the method can obtain the quality features Point matching set that quality is more excellent, robustness is higher and more.On the basis of the basis matrix that the present invention obtains at RANSAC algorithm, use SURF (acceleration robust features) property detector to detect, describe and coupling, then introduce epipolar-line constraint method and reject erroneous matching, thus obtain the coupling set of high-quality more.The present invention is applicable to various view data, such as: large scene, localized target object and there is brightness, block, the image of the change such as rotation.The result obtained can be applicable to three-dimensional reconstruction, target following, in the technical fields such as object identification.
The present invention solves the technical scheme that its technical matters is taked: a kind of acquisition methods of image quality features set of matches, the method is divided into four-stage, comprise: (1) describes the stage in SURF feature detection, find from current image to be matched can with the position of other image matched well, each regioinvertions around the key point detected is become a compacter and stable descriptor (descriptor is used for mating with the descriptor of other images).(2) in the characteristic matching stage, the descriptor treating matching image carries out unique point initial matching according to search strategy.(3) in the basis matrix stage calculating unique point and space projection relation, stochastic sampling unification algorism (RANSAC) is used to obtain high-quality basis matrix.(4) in coupling purification phase, in order to obtain the coupling of reliable high-quality, unique characteristics of image point correspondence is found out.On the basis establishing unique point initial matching, utilize the basis matrix obtained, introduce epipolar-line constraint concept, carry out the rejecting of Mismatching point, obtain the set of high-quality coupling.
Method flow:
Step 1: read two width images to be matched, be defined as left figure and right figure, obtains the Input matching set of the RANSAC algorithm calculating basis matrix;
Step 1-1: the unique point detecting two width images with SURF property detector respectively;
Step 1-2: the unique point descriptor calculating two width images with SURF describer respectively;
Step: 1-3: according to the principle of arest neighbors matching strategy, uses adaptation to carry out bi-directional matching to descriptor.Find each unique point of left figure to two optimum matching of right figure, find two optimum matching of each unique point in left figure in right figure;
Step 1-4: distance rates is tested.Process two coupling set respectively, calculate the distance ratio that Optimum Matching is mated with suboptimum, remove the coupling that distance rates is greater than given threshold value;
Step 1-5: the uniqueness principle utilizing Feature Points Matching right, carries out symmetry test.When index value in two coupling set is symmetrical, extracts this coupling set, reject the set of asymmetric coupling;
Step 2: use RANSAC algorithm to calculate basis matrix.According to polarity geometric projection relation, calculate the basis matrix with maximum Matching supporting set, return the set of high-quality coupling and basis matrix that meet this basis matrix;
Step 3: for two width images to be matched, carry out Feature Points Matching according to descriptor content;
Step 3-1: according to the principle of arest neighbors matching strategy, uses adaptation to find each unique point of left figure to an optimum matching of right figure, returns coupling set;
Step 3-2: according to the principle of arest neighbors matching strategy, uses adaptation to find the optimum matching of each unique point in left figure in right figure, returns coupling set;
Step 4: symmetry test is carried out to two couplings set obtained.Index value in two coupling set retains this coupling set time symmetrical, rejects this coupling set time asymmetric.Return the set of symmetric coupling;
Step 5: the polar curve of computed image unique point in correspondence image;
Step 5-1: according to polar curve definition, calculate the polar curve of all unique points on right figure in left figure with basis matrix;
Step 5-2: according to polar curve definition, calculate the polar curve of all unique points on left figure in right figure with basis matrix;
Step 6: epipolar-line constraint test is carried out to the coupling set tested by symmetry.Reject the coupling of mistake, pass back through the coupling set of epipolar-line constraint test;
Step 6-1: the match point on the left figure that in the set of taking-up coupling, reference key is corresponding, the match point on the right figure that taking-up coupling trains index corresponding in gathering;
Step 6-2: use polar curve function, the match point on left figure is brought into the polar curve function of right figure unique point, calculate its functional value, the match point on right figure is brought into the polar curve function of left figure unique point, calculate its functional value;
Step 6-3: more left figure and right figure on polar curve functional value, functional value get be just simultaneously 0 coupling retain, that is, the coupling of match point on the respective polar curve of correspondence could retain, otherwise rejecting;
Step 7: for by the coupling set of epipolar-line constraint, rejects on same polar curve, to intersect the point of mispairing, return final coupling set;
Step 7-1: the polar curve equation coefficient of more each match point, finds the set of matches of match point on same polar curve;
Step 7-2: to the set of matches on same polar curve, calculates the coordinate position between two between match point on left figure, calculates the coordinate position between two between match point on right figure;
Step 7-3: be multiplied by the coordinate position of two images in left and right, the value obtained is less than 0, direction is contrary, and match point intersects mutually, returns cross one another coupling set index value;
Step 7-4: for the coupling set by epipolar-line constraint, reject the coupling of mutual cross-index value, namely the set of matches of reservation is final high-quality coupling set.
Beneficial effect:
1, the present invention is by introducing the concept of epipolar-line constraint, obtains the high and coupling set that quantity is many of accuracy.
2, the image high-quality coupling acquisition methods that the present invention uses can be applicable to any image.
3, the basis matrix accuracy of the present invention's use is high.
4, the coupling set that the present invention obtains comprises former methodical coupling set completely.
5, computing of the present invention is simple, and processing speed is fast.
Accompanying drawing explanation
Fig. 1 is method flow diagram of the present invention.
Fig. 2 is the epipolar-line constraint schematic diagram that the present invention relates to.
Fig. 3 is design sketch polar curve of the present invention occurring coupling crossing instances.
Fig. 4 is the matching result audio-visual picture of the view data that the present invention uses.
Fig. 5 is the histogram that data result that the present invention carries out testing compares.
Embodiment
For a better understanding of the present invention based on a kind of image quality features set of matches acquisition methods of epipolar-line constraint, below in conjunction with accompanying drawing, the specific embodiment of the present invention is further described.The language used during the example implemented describes does not cause limiting to the claimed invention.
Embodiment one
As shown in Figure 1, the present invention proposes a kind of acquisition methods of image quality features set of matches, the method specifically comprises the following steps:
Step 1: read two width image image1 and image2 to be matched, obtains the Input matching set of the RANSAC algorithm calculating basis matrix;
2D picture point in a visual angle briefly can be mapped to the matrix on the polar curve in another visual angle by basis matrix F (that is: Fundamental Matrix) exactly.It has following features:
(1) basis matrix F to be an order be 23 × 3 matrixes, its degree of freedom is 7;
(2) basis matrix F is unique under difference non-zero constant factor;
(3) basis matrix F can by the corresponding point between image to obtaining;
(4) Given Graph is as the m of in I, and the polar curve l ' in its corresponding image I ' can represent l '=Fm.Similarly, l=Fm '.L represents the polar curve of the some m ' in image I ';
(5) the upper all polar curves of image I (I ') meet at limit e (e '), therefore have: Fe=0, Fe '=0.
Before utilizing epipolar-line constraint, need to obtain basis matrix.The method obtaining basis matrix has a lot, comprising: 7 algorithms, 8 algorithms, FNS, CFNS, M-estimators, LMeds, RANSAC algorithms.Described method choice uses the RANSAC algorithm of robust method to obtain the basis matrix F of high-quality;
The principle of stochastic sampling unification algorism (that is: RANSAC) is in the Feature Points Matching set of input, select 8 (that is: calculating the minimum number of basis matrix) coupling randomly, calculates their basis matrix F (that is: fundamental matrix).Set of matches left in set is used for supporting this basis matrix, is namely the number using RANSAC method to calculate interior point (that is: inlier), and interior point is here the point within prediction Corresponding matching point position ε.The described process of random selection repeats S time.Sample set finally containing point in is at most by as final solution.
The principle mandates of RANSAC algorithm, the coupling set accuracy of algorithm input is higher, and the basis matrix support set of acquisition is larger, thus the quality of basis matrix is higher.Therefore the Input matching set obtaining RANSAC algorithm specifically can be realized by following steps:
Step 1-1: the unique point detecting two width images with SURF property detector respectively;
Use property detector detects the unique point on image to be matched, and such property detector comprises: FAST, SIFT, SURF, ORB, FREAK property detector.The method of the invention choice for use robustness is high, the SURF property detector that detection speed is fast.
When attempt between images matching characteristic time, target object in the picture usually with size, rotate change, in order to address this problem, introduce the acceleration robust features detecting device SURF of size constancy.SURF detects feature and arranges from big to small according to unique point intensity, and the feature that intensity is too little can affect the correctness of subsequent match.Institute thinks raising speed and precision, and we arrange upper limit N to the unique point number looked for, and preferred N is less than or equal to 1200.
Step 1-2: the descriptor calculating unique point on two width images with SURF describer respectively;
Use describer to obtain the descriptor of unique point, such describer comprises: SIFT, SURF, FREAK, ORB describer, the method for the invention choice for use SURF describer.
Descriptor be exactly by the key point detected around each regioinvertions become a compacter and stable descriptor.The feature that the descriptor obtained has is, exists and rotate translation, brightness between (corresponding) image, has better unchangeability, keep the distinction between difference (non-corresponding) image block simultaneously during the apparent change such as shade.Corresponding describer can be used to calculate the descriptor of unique point according to the property detector used.
Step: 1-3: according to the principle of arest neighbors matching strategy, uses adaptation to carry out bi-directional matching to descriptor.Find each unique point of left figure image1 to two optimum matching of right figure image2, find two optimum matching of each unique point in left figure image1 in right figure image2;
The principle of arest neighbors matching strategy (NNDR) compares nearest neighbor distance and time nearest neighbor distance, this nearest neighbor distance be from known with the unmatched piece image of target obtain, we define such nearest neighbor distance ratio and are
Wherein d
1and d
2arest neighbors and time nearest neighbor distance, D
athe sub-D of goal description
band D
cits two nearest neighbours.
Step 1-4: distance rates is tested.Process two coupling set respectively, calculate the distance ratio that Optimum Matching is mated with suboptimum, remove the coupling that distance rates is greater than given threshold value;
Only have arest neighbors to differ certain value with time neighborhood matching distance value, just think that this arest neighbors coupling is that optimum is the most stable, the NNDR distance ratio of the method for the invention choice for use is 0.5f.
Step 1-5: the uniqueness principle utilizing Feature Points Matching right carries out symmetry test.When index value in two coupling set is symmetrical, extracts this coupling set, reject the set of asymmetric coupling;
The uniqueness principle utilizing Feature Points Matching right, the matching double points in two images subject to registration should be one to one, and matching double points p, q relation is as formula 2.Namely require the necessary one_to_one corresponding of the index value in coupling set, the coupling set not meeting one_to_one corresponding condition must be just the coupling of mistake.
{ p → q} ∩ { q → p} formula 2
Step 2: use RANSAC algorithm to calculate basis matrix.According to polarity geometric projection relation, calculate the basis matrix with maximum Matching supporting set, return the set of high-quality coupling and basis matrix that meet this basis matrix;
Step 3: for two width images to be matched, carry out Feature Points Matching according to descriptor content;
Step 3-1: according to the principle of arest neighbors matching strategy, uses adaptation to find each unique point of left figure image1 to an optimum matching of right figure image2, returns coupling set;
Step 3-2: according to the principle of arest neighbors matching strategy, uses adaptation to find the optimum matching of each unique point in left figure image1 in right figure image2, returns coupling set;
Step 4: symmetry test is carried out to two couplings set obtained.Index value in two coupling set retains this coupling set time symmetrical, rejects this coupling set time asymmetric.Return the set of symmetric coupling;
Namely, according to the uniqueness principle of Image Feature Point Matching, filter out erroneous matching in coupling set, check the reference key value in two coupling set and training index value, coupling reservation must meet, 1) the reference key value of left coupling equals the training index value of right coupling; 2) the training index value of left coupling equals the reference key value of right coupling.
As shown in Figure 2 be the schematic diagram of epipolar-line constraint.Observe two camera C of scene point
0and C
1follow the trail of the straight line connecting 3D point and image center, the some p of space 3D point P in piece image can be found
0(x, y).On the contrary, p on image is positioned at
0the point of position can be arranged on the arbitrfary point of this straight line of 3d space, if needed by this picture point p
0find the corresponding point p in another piece image
1, must search for along this straight line at another image surface, this imaginary straight line is called as a p
0polar curve (Epipolar).An end points of polar curve is projected as boundary with the infinite point on raw observation line, and another end points is that former video camera center is projected as boundary at second video camera, and this projection is exactly limit (Epipole).Basis matrix F is by the 2D picture point p in a visual angle
0the polar curve expression formula be mapped in another visual angle is
Therefore picture point p
0match point p
1(x ', y ') should equation be met in correspondence image
Utilize the principle of epipolar-line constraint, the method filtering out high-quality coupling specifically can be realized by following steps:
Step 5: calculate the polar curve of unique point in correspondence image according to formula 3;
Step 5-1: according to polar curve definition, calculate the polar curve of all unique points on right figure in left figure with basis matrix;
Step 5-2: according to polar curve definition, calculate the polar curve of all unique points on left figure in right figure with basis matrix;
Step 6: epipolar-line constraint test is carried out to the coupling set tested by symmetry.Reject the coupling of mistake, pass back through the coupling set of epipolar-line constraint test;
Step 6-1: the match point on the left figure that in the set of taking-up coupling, reference key is corresponding, the match point on the right figure that taking-up coupling trains index corresponding in gathering;
Step 6-2: according to formula 4, obtains the polar curve function of formula 5, the match point on left figure is brought into the polar curve function of right figure unique point, calculates its functional value, the match point on right figure is brought into the polar curve function of left figure unique point, calculates its functional value;
Step 6-3: more left figure and right figure on polar curve functional value, functional value be simultaneously 0 coupling retain, that is, the coupling of match point on the respective polar curve of correspondence could retain, otherwise rejecting.
Wherein the polar curve equation value of step 6-2 and 6-3 calculating and compare and use the polar curve equation of formula 5
Wherein l
1, l
2, l
3for known polar curve coefficient value, current point is brought into the polar curve equation of the point mated with it, when the f (x) only calculated is 0, just represent that point is on polar curve, coupling now retains.
Step 7: for the coupling set by epipolar-line constraint, rejects the coupling that match point intersects on same polar curve, returns final high-quality coupling set;
As shown in Figure 3 be in most cases, there is the matching error of the match point crossover phenomenon that same polar curve occurs.Have two to coupling (k1, k1 ') and (k2, k2 '), wherein k1, k2 are on a polar curve of left figure image1, k1 ', k2 ' and on a polar curve of right figure image2.If formula 6 exists, then matching error is described, two couplings are all removed.
In coupling set, the erroneous matching that Feature Points Matching is intersected is deleted, and obtains final high-quality matching process and specifically can be realized by following steps:
Step 7-1: the polar curve equation coefficient of more each match point, finds the set of matches of match point on same polar curve;
Judge that the process of match point on same polar curve is, 1) for the match point between two on present image, compare three coefficients of polar curve equation respectively, round according to the principle rounded up; 2), when only having three coefficients of a point and the polar curve equation of another point simultaneously all equal, just satisfied two match points are on same polar curve.
Step 7-2: to the set of matches on same polar curve, calculates the coordinate position between two between match point on left figure, calculates the coordinate position between two between match point on right figure;
Step 7-3: be multiplied by the coordinate position of two images in left and right, the value obtained is less than 0, direction is contrary, and match point intersects mutually, returns cross one another coupling set index value;
Step 7-4: for the coupling set by epipolar-line constraint, reject the coupling of mutual cross-index value, namely the set of matches of reservation is final high-quality coupling set;
So far, the present invention completes the screening feature high-quality set of matches method based on epipolar-line constraint, obtains the set of more high-quality coupling.
Embodiment two
Superiority for a more clear understanding of the present invention, concrete steps in conjunction with the embodiments, below list the present invention and existing former method uses church, canal, during circlebuild, monument and parliament five groups of view data, the comparative result in Feature Points Matching.
As shown in Table 1, use in church view data, the high-quality number of matches that the former method of old obtains is 15, and new new method is 99, and the high-quality number of matches that new method obtains is 6.6 times of former method.The including item that new method comprises the high-quality coupling number of aging method is 15, comprises the high-quality set of matches of aging method completely.New method is much higher than the high-quality number of matches that aging method obtains, and new method is the high-quality set of matches comprising aging method acquisition completely.In new method, the erroneous matching crossing mono-of the mutual intersection that polar curve exists has 4 in parliament data, and the high-quality coupling number obtained after rejecting is 189, is 2.1 times of former method.
Table 1: the comparative result of the different images data of new algorithm and old algorithm
Note: old represents former method high-quality coupling number; New represents the high-quality coupling number that new method obtains; Including is the former algorithm high-quality number of matches that new algorithm comprises; In new algorithm, polar curve there is the erroneous matching quantity of intersecting mutually in crossing.
Fig. 4 is the matching result figure of five groups of view data that the present invention uses.Each white line represents a pair outstanding coupling.Shown in each group image left side is the high-quality coupling figure that former method obtains, and shown in the right is the high-quality coupling figure that new method of the present invention obtains.Can significantly find out from matching effect figure, it is obviously more than the high-quality coupling number that left primitive method obtains that the right figure of each group mates number.Coupling in experimental verification new method is the high-quality coupling comprising former method completely.Therefore can reach a conclusion, the result that the method that the present invention uses obtains is better.
Fig. 5 is the present invention and the existing method histogram at image quality features Point matching collection quantitative aspects comparative result, and namely in table 1, old and new two of five groups of data arranges.The row of chart is five groups of data, and row are high-quality number of matches.Can be found out intuitively by figure, for different view data, the high-quality coupling number that the present invention obtains is apparently higher than existing methodical result.
Claims (10)
1. an acquisition methods for image quality features set of matches, is characterized in that, described method comprises the steps:
Step 1: read two width images to be matched, obtains the Input matching set of the RANSAC algorithm calculating basis matrix;
Step 2: use RANSAC algorithm to calculate basis matrix, according to polarity geometric projection relation, calculate the basis matrix with maximum Matching supporting set, returns the set of high-quality coupling and basis matrix that meet this basis matrix;
Step 3: for two width images to be matched, carry out Feature Points Matching according to descriptor content;
Step 4: carry out symmetry test to two couplings set obtained, retains this coupling set when the index value in two coupling set is symmetrical, reject this coupling set, return the set of symmetric coupling time asymmetric;
Step 5: the polar curve of computed image unique point in correspondence image;
Step 6: carry out epipolar-line constraint test to the coupling set tested by symmetry, rejects the coupling of mistake, passes back through the coupling set of epipolar-line constraint test;
Step 7: for the coupling set by epipolar-line constraint, rejects the coupling that match point intersects on same polar curve, returns final high-quality coupling set.
2. the acquisition methods of a kind of image quality features set of matches according to claim 1, is characterized in that, described step 1 is the process of the Input matching set obtaining the RANSAC algorithm calculating basis matrix, comprises the steps:
Step 1-1: the unique point detecting two width images with SURF property detector respectively;
Described step 1-1 use property detector detects the unique point on image to be matched, and such property detector comprises: FAST, SIFT, SURF, ORB, FREAK property detector; Described method choice uses robustness high, the SURF property detector that detection speed is fast;
Described step 1-1 is the unique point detecting image to be matched with SURF property detector, and for improving speed and precision, arrange upper limit N to the unique point number looked for, preferred N is less than or equal to 1200;
Step 1-2: the descriptor calculating unique point on two width images with SURF describer respectively;
Described step 1-2 uses describer to obtain the descriptor of unique point, and such describer comprises: SIFT, SURF, FREAK, ORB describer, and described method uses SURF describer;
Step: 1-3: according to the principle of arest neighbors matching strategy, uses adaptation to carry out bi-directional matching to descriptor, finds each unique point of left figure to two optimum matching of right figure, find two optimum matching of each unique point in left figure in right figure;
Step 1-4: distance rates is tested; Process two coupling set respectively, calculate the distance ratio that Optimum Matching is mated with suboptimum, remove the coupling that distance rates is greater than given threshold value;
The NNDR distance ratio that described method choice uses is 0.5f;
Step 1-5: the uniqueness principle utilizing Feature Points Matching right carries out symmetry test, when the index value in two coupling set is symmetrical, extracts this coupling set, rejects the set of asymmetric coupling.
3. the acquisition methods of a kind of image quality features set of matches according to claim 1, is characterized in that, described step 2 uses RANSAC algorithm, obtains the process of basis matrix;
The method of described acquisition basis matrix, comprising: 7 algorithms, 8 algorithms, FNS, CFNS, M-estimators, LMeds, RANSAC algorithms; Described method choice uses the RANSAC algorithm of robust method to obtain the basis matrix F of high-quality.
4. the acquisition methods of a kind of image quality features set of matches according to claim 1, is characterized in that, described step 3 is the processes obtaining Image Feature Matching set to be matched, comprises the steps:
Step 3-1: according to the principle of arest neighbors matching strategy, uses adaptation to find each unique point of left figure to an optimum matching of right figure, returns coupling set;
Step 3-2: according to the principle of arest neighbors matching strategy, uses adaptation to find the optimum matching of each unique point in left figure in right figure, returns coupling set.
5. the acquisition methods of a kind of image quality features set of matches according to claim 1, it is characterized in that, described step 4 is the uniqueness principles according to Image Feature Point Matching, filters out the process of erroneous matching in coupling set;
Described step 4 carries out the concrete steps that symmetry filters out erroneous matching:
Check the reference key value in two coupling set and training index value, coupling reservation must meet, 1) the reference key value of left coupling equals the training index value of right coupling; 2) the training index value of left coupling equals the reference key value of right coupling.
6. the acquisition methods of a kind of image quality features set of matches according to claim 1, is characterized in that, described step 5 calculates the polar curve process of unique point in correspondence image, comprises the steps:
Step 5-1: according to polar curve definition, calculate the polar curve of all unique points on right figure in left figure with basis matrix;
Step 5-2: according to polar curve definition, calculate the polar curve of all unique points on left figure in right figure with basis matrix.
7. the acquisition methods of a kind of image quality features set of matches according to claim 1, is characterized in that, described step 6 is the processes coupling set by symmetrical test being used to epipolar-line constraint, comprises the steps:
Step 6-1: the match point on the left figure that in the set of taking-up coupling, reference key is corresponding, the match point on the right figure that taking-up coupling trains index corresponding in gathering;
Step 6-2: use polar curve function, the match point on left figure is brought into the polar curve function of right figure unique point, calculate its functional value, the match point on right figure is brought into the polar curve function of left figure unique point, calculate its functional value;
Step 6-3: more left figure and right figure on polar curve functional value, functional value be simultaneously 0 coupling retain, that is, the coupling of match point on the respective polar curve of correspondence could retain, otherwise rejecting;
The calculating of polar curve equation value and the concrete steps of comparison procedure of described step 6-2 and 6-3 are:
Use polar curve equation is
Wherein l
1, l
2, l
3for known polar curve coefficient value, current point is brought into the polar curve equation of the point mated with it, when the f (x) only calculated is 0, just represent that point is on polar curve, coupling now retains.
8. the acquisition methods of a kind of image quality features set of matches according to claim 1, is characterized in that, described step 7 is the processes obtaining final high-quality set of matches result, comprises the steps:
Step 7-1: the polar curve equation coefficient of more each match point, finds the set of matches of match point on same polar curve;
Step 7-1 judges that the concrete steps of the process of match point on same polar curve are:
For the match point between two on present image, 1) compare three coefficients of polar curve equation respectively, round according to the principle rounded up; 2), when only having three coefficients of a point and the polar curve equation of another point simultaneously all equal, just satisfied two match points are on same polar curve;
Step 7-2: to the set of matches on same polar curve, calculates the coordinate position between two between match point on left figure, calculates the coordinate position between two between match point on right figure;
Step 7-3: be multiplied by the coordinate position of two images in left and right, the value obtained is less than 0, direction is contrary, and match point intersects mutually, returns cross one another coupling set index value;
Step 7-4: for the coupling set by epipolar-line constraint, reject the coupling of mutual cross-index value, namely the set of matches of reservation is final high-quality coupling set.
9. the acquisition methods of a kind of image quality features set of matches according to claim 1, it is characterized in that, the four-stage of described method, comprise: (1) describes the stage in SURF feature detection, from current image to be matched find can with the position of other image matched well, each regioinvertions around the key point detected is become a compacter and stable descriptor, that is: descriptor is used for and the descriptor of other images mates; (2) in the characteristic matching stage, the descriptor treating matching image carries out unique point initial matching according to search strategy; (3) in the basis matrix stage calculating unique point and space projection relation, RANSAC algorithm is used to obtain high-quality basis matrix; (4) in coupling purification phase, in order to obtain the coupling of reliable high-quality, find out unique characteristics of image point correspondence, on the basis establishing unique point initial matching, utilize the basis matrix obtained, introduce epipolar-line constraint concept, carry out the rejecting of Mismatching point, obtain the set of high-quality coupling.
10. the acquisition methods of a kind of image quality features set of matches according to claim 1 is applied to three-dimensional reconstruction, target following, object recognition technique field.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510010357.2A CN104680516B (en) | 2015-01-08 | 2015-01-08 | A kind of acquisition methods of image quality features set of matches |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510010357.2A CN104680516B (en) | 2015-01-08 | 2015-01-08 | A kind of acquisition methods of image quality features set of matches |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104680516A true CN104680516A (en) | 2015-06-03 |
| CN104680516B CN104680516B (en) | 2017-09-29 |
Family
ID=53315509
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510010357.2A Expired - Fee Related CN104680516B (en) | 2015-01-08 | 2015-01-08 | A kind of acquisition methods of image quality features set of matches |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104680516B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105245841A (en) * | 2015-10-08 | 2016-01-13 | 北京工业大学 | CUDA (Compute Unified Device Architecture)-based panoramic video monitoring system |
| CN107481269A (en) * | 2017-08-08 | 2017-12-15 | 西安科技大学 | A kind of mine multi-cam moving target continuous tracking method |
| CN107993193A (en) * | 2017-09-21 | 2018-05-04 | 沈阳工业大学 | The tunnel-liner image split-joint method of surf algorithms is equalized and improved based on illumination |
| CN108615248A (en) * | 2018-04-27 | 2018-10-02 | 腾讯科技（深圳）有限公司 | Method for relocating, device, equipment and the storage medium of camera posture tracing process |
| TWI677852B (en) * | 2017-07-20 | 2019-11-21 | 大陸商北京三快在線科技有限公司 | A method and apparatus, electronic equipment, computer readable storage medium for extracting image feature |
| CN110874818A (en) * | 2018-08-31 | 2020-03-10 | 阿里巴巴集团控股有限公司 | Image processing and virtual space construction method, device, system and storage medium |
| CN111028284A (en) * | 2019-10-31 | 2020-04-17 | 浙江未来技术研究院(嘉兴) | Binocular vision stereo matching method and device based on homonymous mark points |
| CN111241979A (en) * | 2020-01-07 | 2020-06-05 | 浙江科技学院 | Real-time obstacle detection method based on image feature calibration |
| CN112016610A (en) * | 2020-08-25 | 2020-12-01 | 济南大学 | Image feature matching method and system |
| CN113223184A (en) * | 2021-05-26 | 2021-08-06 | 北京奇艺世纪科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN113298720A (en) * | 2021-04-21 | 2021-08-24 | 重庆邮电大学 | Self-adaptive overlapped image rotation method |
| CN114998773A (en) * | 2022-08-08 | 2022-09-02 | 四川腾盾科技有限公司 | Characteristic mismatching elimination method and system suitable for aerial image of unmanned aerial vehicle system |
| CN117115708A (en) * | 2023-08-24 | 2023-11-24 | 中海建筑有限公司 | Foggy environment bridge cable force identification method based on deep learning micro-motion amplification technology |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101350101A (en) * | 2008-09-09 | 2009-01-21 | 北京航空航天大学 | Method for auto-registration of multi-amplitude deepness image |
| CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
| US20140037189A1 (en) * | 2012-08-02 | 2014-02-06 | Qualcomm Incorporated | Fast 3-D point cloud generation on mobile devices |
| CN104240267A (en) * | 2014-09-19 | 2014-12-24 | 南京理工大学 | Moving object detection method based on three-dimensional structure coordinate constraint |
-
2015
- 2015-01-08 CN CN201510010357.2A patent/CN104680516B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101350101A (en) * | 2008-09-09 | 2009-01-21 | 北京航空航天大学 | Method for auto-registration of multi-amplitude deepness image |
| CN102722731A (en) * | 2012-05-28 | 2012-10-10 | 南京航空航天大学 | Efficient image matching method based on improved scale invariant feature transform (SIFT) algorithm |
| US20140037189A1 (en) * | 2012-08-02 | 2014-02-06 | Qualcomm Incorporated | Fast 3-D point cloud generation on mobile devices |
| CN104240267A (en) * | 2014-09-19 | 2014-12-24 | 南京理工大学 | Moving object detection method based on three-dimensional structure coordinate constraint |
Non-Patent Citations (4)
| Title |
|---|
| 向登宁 等: "利用极线约束方法实现图像特征点的匹配", 《北京机械工业学院学报》 * |
| 喻夏琼 等: "运动序列中动目标检测的稳健性方法", 《激光与光电子学进展》 * |
| 张培耘 等: "基于RANSAC算法的极线约束立体视觉匹配方法研究", 《组合机床与自动化加工技术》 * |
| 时颢 等: "基于SURF与BBF的棉花图像匹配算法", 《江苏农业科学》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105245841A (en) * | 2015-10-08 | 2016-01-13 | 北京工业大学 | CUDA (Compute Unified Device Architecture)-based panoramic video monitoring system |
| CN105245841B (en) * | 2015-10-08 | 2018-10-09 | 北京工业大学 | A kind of panoramic video monitoring system based on CUDA |
| TWI677852B (en) * | 2017-07-20 | 2019-11-21 | 大陸商北京三快在線科技有限公司 | A method and apparatus, electronic equipment, computer readable storage medium for extracting image feature |
| US11282295B2 (en) | 2017-07-20 | 2022-03-22 | Beijing Sankuai Online Technology Co., Ltd | Image feature acquisition |
| CN107481269A (en) * | 2017-08-08 | 2017-12-15 | 西安科技大学 | A kind of mine multi-cam moving target continuous tracking method |
| CN107993193B (en) * | 2017-09-21 | 2021-06-11 | 沈阳工业大学 | Tunnel lining image splicing method based on illumination equalization and surf algorithm improvement |
| CN107993193A (en) * | 2017-09-21 | 2018-05-04 | 沈阳工业大学 | The tunnel-liner image split-joint method of surf algorithms is equalized and improved based on illumination |
| CN108615248B (en) * | 2018-04-27 | 2022-04-05 | 腾讯科技（深圳）有限公司 | Method, device and equipment for relocating camera attitude tracking process and storage medium |
| US11481923B2 (en) | 2018-04-27 | 2022-10-25 | Tencent Technology (Shenzhen) Company Limited | Relocalization method and apparatus in camera pose tracking process, device, and storage medium |
| CN108615248A (en) * | 2018-04-27 | 2018-10-02 | 腾讯科技（深圳）有限公司 | Method for relocating, device, equipment and the storage medium of camera posture tracing process |
| CN110874818A (en) * | 2018-08-31 | 2020-03-10 | 阿里巴巴集团控股有限公司 | Image processing and virtual space construction method, device, system and storage medium |
| CN110874818B (en) * | 2018-08-31 | 2023-06-23 | 阿里巴巴集团控股有限公司 | Image processing and virtual space construction method, device, system and storage medium |
| CN111028284A (en) * | 2019-10-31 | 2020-04-17 | 浙江未来技术研究院(嘉兴) | Binocular vision stereo matching method and device based on homonymous mark points |
| CN111241979B (en) * | 2020-01-07 | 2023-06-23 | 浙江科技学院 | Real-time obstacle detection method based on image feature calibration |
| CN111241979A (en) * | 2020-01-07 | 2020-06-05 | 浙江科技学院 | Real-time obstacle detection method based on image feature calibration |
| CN112016610B (en) * | 2020-08-25 | 2022-05-31 | 济南大学 | Image feature matching method and system |
| CN112016610A (en) * | 2020-08-25 | 2020-12-01 | 济南大学 | Image feature matching method and system |
| CN113298720B (en) * | 2021-04-21 | 2022-08-19 | 重庆邮电大学 | Self-adaptive overlapped image rotation method |
| CN113298720A (en) * | 2021-04-21 | 2021-08-24 | 重庆邮电大学 | Self-adaptive overlapped image rotation method |
| CN113223184A (en) * | 2021-05-26 | 2021-08-06 | 北京奇艺世纪科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN113223184B (en) * | 2021-05-26 | 2023-09-05 | 北京奇艺世纪科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN114998773A (en) * | 2022-08-08 | 2022-09-02 | 四川腾盾科技有限公司 | Characteristic mismatching elimination method and system suitable for aerial image of unmanned aerial vehicle system |
| CN117115708A (en) * | 2023-08-24 | 2023-11-24 | 中海建筑有限公司 | Foggy environment bridge cable force identification method based on deep learning micro-motion amplification technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104680516B (en) | 2017-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104680516B (en) | A kind of acquisition methods of image quality features set of matches | |
| Revaud et al. | Epicflow: Edge-preserving interpolation of correspondences for optical flow | |
| US8929602B2 (en) | Component based correspondence matching for reconstructing cables | |
| CN101216895B (en) | An automatic extracting method for ellipse image features in complex background images | |
| JP3242529B2 (en) | Stereo image matching method and stereo image parallax measurement method | |
| Tsai et al. | Block-based vanishing line and vanishing point detection for 3D scene reconstruction | |
| CN108596975A (en) | A kind of Stereo Matching Algorithm for weak texture region | |
| CN106530407A (en) | Three-dimensional panoramic splicing method, device and system for virtual reality | |
| JPH0719832A (en) | Extracting method for corresponding points of pulirity of images | |
| CN104182968A (en) | Method for segmenting fuzzy moving targets by wide-baseline multi-array optical detection system | |
| CN108550166A (en) | A kind of spatial target images matching process | |
| CN103927785A (en) | Feature point matching method for close-range shot stereoscopic image | |
| CN107808140A (en) | A kind of monocular vision Road Recognition Algorithm based on image co-registration | |
| Nasrabadi et al. | Stereo vision correspondence using a multichannel graph matching technique | |
| US9087381B2 (en) | Method and apparatus for building surface representations of 3D objects from stereo images | |
| KR102577663B1 (en) | Image pose estimation system using feature point matching pair filtering and image pose estimation method using thereof | |
| Kallasi et al. | Computer vision in underwater environments: A multiscale graph segmentation approach | |
| Abd Gani et al. | Depth Map Information from Stereo Image Pairs using Deep Learning and Bilateral Filter for Machine Vision Application | |
| Pugeault et al. | Multi-Modal Matching Applied to Stereo. | |
| CN114199205B (en) | Binocular Ranging Method Based on Improved Quadtree ORB Algorithm | |
| CN102567992A (en) | Image matching method of occluded area | |
| Luo et al. | Stereo matching and occlusion detection with integrity and illusion sensitivity | |
| CN106056599B (en) | A kind of object recognition algorithm and device based on Object Depth data | |
| CN105631887B (en) | Based on the adaptive two step parallax modification methods and system for supporting weight matching algorithm | |
| Lyra et al. | Development of an efficient 3D reconstruction solution from permissive open-source code |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170929 |